It's actually insane that there are papers coming out about AI capabilities evaluations using two-year-old models. We have to reduce the length of time to get studies published or academic publishing will have to become completely irrelevant for AI.